Constrained Text Generation on Natural Language Benchmarks Between (ubd.) constraint
93.8SaturationDIFFINITY
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DIFFINITYGuidance scale=2.52026.02 | 93.8 | 100 | 57.4 | 33.3 | |
| GPT2-Large-GCDGuided Character Decoding=true2026.02 | 3.5 | 10 | 71.2 | 47.1 | |
| GPT2-Medium-GCDGuided Character Decoding=true2026.02 | 0.2 | 5 | 132.6 | 33.3 |